Figure | Pixabay
A person's actual age does not necessarily faithfully reflect the degree of aging of the body. But the eyes may be able to...
Written by | chestnut
Review | Clefable
A person's chronological age is calculated from the day of birth.
However, being born for as long does not mean that the body is as old. Some people are 40 years old, and their physical functions can be comparable to ordinary people in their 20s, and some people are 40 years old, and the various organs run like ordinary people around 60 years old.
If the actual age is set aside, the degree of aging of the human body can also be observed through other means. For example, the overall decline in the level of DNA methylation in the genome is a manifestation of the body's aging. As a result, people often think of the degree of DNA methylation as an "aging clock."
The state of the retina can be used to predict age | VMR Institute
Now, the eye can also be a window into the aging process and can even be used to assess the risk of death. Recently, scientists have found that people who are older than their actual age are more likely to die in recent years than others. The research team published the results in the British Journal of Ophthalmology.
But how do scientists tell how old a person's retina is?
Ai can see the age of the retina
When a person slowly grows old, the operating state of various organs in the body will change, from vigorous to disrepair, and the eyes are no exception. If you analyze the fundus images, you may find that people of different ages have different retinal faces.
With this in mind, scientists from the Guangdong Academy of Medical Sciences, the University of Melbourne and other institutions have developed a deep learning AI that wants to predict the age of the owner based on what the retina looks in the fundus image.
To this end, the research team obtained 19 200 fundus maps from the UK Biobank. The images were taken from 11,052 people, ranging in age from 40 to 69 years, but had one thing in common: none had a history of prior medical illness before the images were taken. Scientists hope that AI can learn how the retina changes with age in people in good health.
AI learns to predict the | based on the actual age of the fundus and the owner of the picture Original paper
Among the more than 10,000 fundus maps, most of the pictures are fed to the AI as training data, and the actual age is marked for the AI to explore the rules. After the training is completed, the remaining pictures will be entered to the AI as verification data (equivalent to exam questions), without marking the actual age, and let the AI predict the age based on the picture.
The team found that the results of the AI predictions were very close to the actual age of the picture owner, with an error of no more than 3.55 years. This not only means that AI training is effective, but also means that the retinal information contained in the image can largely reflect the actual age of humans. In other words, the age of the retina of a healthy person is more consistent with the actual age.
Of course, learning the data of these healthy people is only to make a "yardstick" to measure the situation of others. And scientists are more concerned about those who are less physically healthy, especially those whose retinas are older than their actual age.
How much is known about the risk of death
After training, the AI that has also passed the acceptance test should observe the fundus map of more people and give age prediction. This time, the image is still from the British Biobank, but it is no longer necessary to screen out healthy people, but whether there is a history of illness or not. The scientists collected eye-eye maps of 35,913 people and also knew how old they were when the pictures were taken and their health (including life and death) for 11 years after the shooting was completed.
The team gave these fundus maps to the AI to predict, and then subtracted the actual age of the retina predicted by the AI to get the age difference. It turned out that the greater the difference, the higher the probability that the owner of the picture would die in the 11 years after the picture was taken.
The difference between the age of the retina predicted by AI minus the actual age can be divided into four groups, each group containing 25% of people. The first group (blue) has the smallest difference and the fourth group (red) has the largest difference. The greater the difference, the smaller the probability of survival (data on the vertical axis) | Original paper
Specifically, for every year that the difference between retinal age and chronological age is widened, the probability of death of the picture owner over an 11-year observation period increases by 2%. In other words, the older the retina is than the actual age, the higher the probability of death and the smaller the chance of survival. Among them, people whose retina is more than 3 years older than their actual age are called "fast ager" by scientists.
Of the more than 30,000 people surveyed, a total of 1,871 died during the observation period. Of these, 321 (17.2%) of deaths could be attributed to cardiovascular disease, 1018 (54.4%) were cancer, and the remaining 532 (28.4%) were other causes. These situations can be discussed separately.
If you look beyond cardiovascular disease and cancer and look at the rest, the difference between retinal age and actual age increases by 3% for each year of widening. If you rank the difference from smallest to largest, the bottom 25 percent of people are 67 percent more likely to die from "other causes" than those in the top 25 percent.
When looking at the case of cardiovascular disease alone, scientists found that the difference between retinal age and actual age and mortality rate was not obvious. But the team believes that this does not mean that the retina cannot reflect the cardiovascular aging process, but with the development of medicine, some once fatal cardiovascular problems can be saved, so that the mortality rate of related diseases has decreased.
Scientists believe that the difference between retinal age and actual age can be used as a biomarker for human aging and can also be used to predict the risk of death.
Is there a scientific basis for this?
If only the results predicted by AI are lacking scientific explanations, it may be difficult to go further by using retinal pictures to estimate the degree of aging. So, what exactly can the retina in the image show?
Kidney disease, for example, is often associated with eye disease. After all, the eye and kidneys have many similarities in structure, development, and genetic pathways: the choroids and glomeruli in the eye have structurally similar networks of blood vessels; the retina and glomeruli filter barriers have similar developmental pathways.
Previous studies have shown that chronic kidney disease is associated with macular degeneration, and studies have shown that abnormal retinal microvascular parameters can be used to predict chronic kidney disease. Emboli in the retina, the clumps that can block blood vessels, have also been shown by scientists to be associated with many diseases such as chronic kidney disease.
The protagonist of this article, the AI that predicts the age of the retina, has a attention mechanism: it will determine which areas on the picture are more conducive to its prediction, and then increase the weight of those parts. The team said that the area of more attention of the AI in the fundus map is also the vascular part of the retina.
The left is the fundus map, and the right is the same fundus map that has been highlighted by AI to highlight the key areas | Original paper
As for cardiovascular disease, although scientists have not found a link between mortality and retinal age difference this time, if the death data is not discussed, there have been many studies that believe that the eye is a window to cardiovascular disease.
For example, Professor Ilias Georgalas of the University of Athens introduced a case. The right eye of a 77-year-old man in Greece developed three brief blurred visions in about 1 hour, each lasting about 5 minutes.
After the visit, the doctor said that his intraocular pressure was normal and his vision was good, but on closer inspection, it was a blood clot that blocked the blood supply to a branch artery of the retina. Looking at the eyes is a simple way for doctors to examine the vascular system in a patient's body. Professor Georgias said that when there are blood vessel problems in other parts, most of them can be seen in the eyes.
For the man, doctors also started with an abnormality in the retina and found that 80% of the blood flow in his right carotid artery was blocked due to atherosclerosis, which was a red flag for stroke. So the man underwent surgery to remove the plaques that had accumulated in the blood vessels. As a physician involved in the treatment, Professor Georgias also collaborated with colleagues to publish a study highlighting the high probability of severe (or even fatal) stroke in patients with blocked retinal branch arteries.
In recent years, scientists have used machine learning AI to explore the link between retinal images and these health problems, both for kidney and cardiovascular diseases. In contrast, studying aging in humans may be more complex than studying specific diseases, and it is difficult to establish an indicator until there is an "aging clock" that relies on DNA methylation to predict age. In 2020, it was the machine learning algorithm DeepMAge that gave the "aging clock" unprecedented accuracy: the margin of error for predicting age was within 3 years.
Now, with the help of AI, scientists use the retina in the picture to predict age, so as to judge the speed of aging and the risk of death. Perhaps someone in the future will also get timely treatment because of AI predictions, so that life can be better extended.
Original paper
https://bjo.bmj.com/content/early/2021/11/17/bjophthalmol-2021-319807
bibliography
https://www.sciencedirect.com/science/article/pii/S0047637408002820
https://medicalxpress.com/news/2022-01-difference-retina-biological-age-person.html
https://www.sciencedirect.com/science/article/pii/S008525381556362X
https://www.thelancet.com/journals/landig/article/PIIS2589-7500(20)30063-7/fulltext
https://www.nejm.org/doi/10.1056/NEJMicm1500779
https://www.livescience.com/52922-eyes-problems-signal-cardiovascular-disease.html
https://www.cell.com/molecular-cell/fulltext/S1097-2765(12)00893-3
http://www.aginganddisease.org/EN/10.14336/AD.2020.1202
This article is reprinted from Universal Science (ID: huanqiukexue) with permission, please contact the original author for secondary reprinting.